Distributed Relational Activation in Large Language Models: A Dimension-Specific Model of Interaction Modulation

This paper introduces the Distributed Relational Activation (DRA) model, a dimension-specific framework for understanding how relational signals influence the behavior of large language models (LLMs). Moving beyond traditional prompt engineering, the study investigates how interaction structure — rather than prompt content alone — shapes the depth, organization, and orientation of generated outputs. Using a controlled qualitative experimental design, identical informational tasks are tested across systematically varied relational conditions, including baseline, minimal, and fully structured configurations, as well as isolated relational signals such as role attribution, collaborative framing, mechanistic constraints, and sequential structuring. The results demonstrate that relational signals do not act as a unified trigger. Instead, they selectively activate distinct dimensions of the generative process, including cognitive framing, interactional engagement, analytical depth, and structural organization. Based on these findings, the paper proposes the DRA model and integrates it with the Relational Modulation Architecture (RMA), offering a structured framework for analyzing and designing human–AI interaction. This work contributes a replicable experimental protocol, a formalized theoretical model, and a foundation for the emerging field of interaction engineering in large language models.